miheerdew/cbce

Finding communities in a bipartite correlation network

cbce's Issues

Generate optional diagnostics information

This could be a pdf document (based on a switch), which give details about how the Bimodules were generated.

Call them "communities" instead of "bimodules"

cbce/R/interaction.R

Line 150 in eb7ff8a

tcltk::tclvalue(env$statusLabel) <- "Filtering bimodules. Please wait."

Add NIH credits to webpage

Clipboard doesn't work when using GUI on Windows

This is because of tkrplot, which copies to the clipboard whenever a new plot is updated.

056aa20 tried to mitigate this by changing the refresh duration to 5 sec, but that obviously didn't make the problem go away.

Better progress estimation in the command line interface

Randomize the order of extractions and use the progress package that gives completion time estimates.

Program slow/freezes for large alpha

On running the program on my data set (100x10K and 100x100k matrices), for alpha = 0.05 the runs are very slow and freeze, while for alpha=0.01 the runs are fine.

Pass correlations from cor-backend to Summary P-val

cbce/R/cbce.R

Line 266 in f3cc6b0

log.pval <- summary_pval(X[, B0$x, drop=FALSE],

Currently summary-pval recalculates the correlations. Could this be causing a significant slowdown for large bimodules? Passing the chached correlations from the backend might help.

Check the input for outliers

At the beginning of the method, raise a warning if the X and Y matrices have any columns which are like 4SD away from the mean. This causes problems with the correlations.

Actually, at least one of the columns need to look like a normal.

Use actual numbering instead of global numbers in extraction index

cbce/R/cbce.R

Line 250 in eb7ff8a

return(c(list("indx" = indx,

Currently extraction index returns a global number and its difficult for the users to make sense of this. Instead use the actual numbering and point out if it were an X or Y index. Perhaps by naming the index.

Randomly chosen starts

If we assume that we are interested in some larger bimodules which can be obtained many times. We can use this to our advantage by sampling randomly. For instance, sampling each node with a .25 percent chance will ensure that we catch most of them.

Include code for setting alpha

Provide code for setting alpha with cross validation.

Finding same communities again

Sometimes communities grow very large and the same set is found again and again. One huersitic to deal with this is to not start within a bimodule that is found.

Pressing browser takes a long time to stop

I think the browser tries to print the name of the arguments, and when that is huge, it takes a long time.

Handling covariates

Like matrix eQTL take the covariates as a new argument, residualize the data, and change the effective sample size as necessary.